System Design: Data Storage at Scale

Master blob storage, distributed file systems, storage paradigm comparisons, Data Lakes vs Warehouses, temperature tiering, and media storage pipelines.

1. Object Storage vs Block Storage vs File Storage

Modern cloud infrastructure provides three distinct storage paradigms tailored for different read/write access patterns and file structure requirements.

Object Storage vs Block Storage vs File Storage Architecture Topologies

System Design

Comparing flat key-value object namespaces, low-latency VM blocks, and shared POSIX file directories

100%
Loading system design canvas…

Object Storage (AWS S3, Google Cloud Storage)

Stores data as immutable objects in a flat namespace. Each object contains payload data, a unique key ID, and custom metadata tags. Accessed over HTTP REST APIs.

Block Storage (AWS EBS, SAN)

Exposes raw unformatted block volumes formatted with traditional file systems (ext4, NTFS) and attached directly to Virtual Machines over high-speed NVMe buses. Ideal for databases.

File Storage (AWS EFS, Network Attached Storage NFS/SMB)

Provides shared hierarchical directory structures with POSIX file locking permissions, accessible concurrently by hundreds of application server instances.

2. Blob Storage & Distributed File Systems (HDFS, GFS)

Distributed file systems (HDFS, GFS) split massive petabyte files into fixed-size block chunks (e.g. 128 MB) and distribute them across a cluster of commodity hardware servers with automatic 3x replication factor.

Distributed File System (HDFS NameNode & DataNode Replication)

System Design

Tracing NameNode metadata block mappings and DataNode 3x block replication for fault tolerance

100%
Loading system design canvas…

NameNode (Master Node)

Maintains in-memory metadata mappings of file paths to block IDs and tracks which physical DataNode servers host each block replica.

DataNodes (Worker Nodes)

Stores raw block chunks on local disk drives and streams data blocks directly to client applications without passing data payloads through the NameNode.

Review how the HDFS Chunk Simulator below calculates block locations and 3x replica node placement.

HDFS Distributed Chunk Routing & Block Replication Simulator

Splitting file streams into fixed-size block chunks and distributing 3x replicas across cluster DataNodes

3. Data Lakes, Warehouses & Temperature Storage Tiering

Enterprise data architectures balance analytics query performance against long-term storage costs using tiered storage lifecycles.

Data Lakes vs Data Warehouses & Temperature Storage Tiering

System Design

Comparing schema-on-read Data Lakes against schema-on-write Warehouses and Hot/Warm/Cold storage costs

100%
Loading system design canvas…

Data Warehouses (Snowflake, BigQuery, Redshift)

Optimized for structured SQL analytics queries. Uses a schema-on-write approach where data is transformed (ETL) before ingestion.

Data Lakes (AWS S3 + Glue, Apache Iceberg)

Stores raw structured, semi-structured (JSON, Parquet), and unstructured (images, audio) data in native format using a flexible schema-on-read approach.

Hot vs Warm vs Cold Storage Lifecycle

Hot Tier (Standard S3): Instant millisecond access for active data. Warm Tier (S3 IA): Lower storage cost for data accessed once per month. Cold Tier (Glacier): Ultra-low cost for long-term archives with 1-12 hour retrieval times.

4. Content Storage and Retrieval for Media Systems

High-throughput media platforms (YouTube, Netflix) decouple file upload ingestion from video streaming delivery. Clients request temporary presigned URLs from API servers, enabling direct browser-to-S3 uploads that bypass application servers entirely.

Media Storage, Transcoding Pipeline & CDN Delivery Architecture

System Design

Tracing presigned URL uploads, asynchronous HLS transcoding, and global CDN edge distribution

100%
Loading system design canvas…

Review the Node.js S3 Presigned URL generator and Multipart Upload manager below to see direct cloud storage uploads in practice.

S3 Presigned URL Upload & Multipart Chunked Upload Manager

Generating secure presigned S3 URLs for direct client uploads and managing multipart chunk completions

Data Storage at Scale Knowledge Verification

1. What defines Object Storage like AWS S3 or Google Cloud Storage?

2. How does HDFS achieve fault tolerance across distributed clusters?

3. What characterizes a Data Lake compared to a Data Warehouse?

4. What is the main trade-off of Cold Storage (Glacier)?

5. Why do media systems generate presigned URLs for object uploads?